Optic Disc and Exudates Segmentation on Retinal Fundus Images Using Mask R-CNN

Diabetic retinopathy (DR) is an illness caused by diabetes. In 2019, the Centers for Disease Control and Prevention (CDC) approximated 1.4 million new illnesses of diabetes. To identify the DR severity, the ophthalmologist checked the appearance of clinical features, and one of them is exudates. Exudates arise due to the accumulation of lipids and proteins due to a ruptured retinal vein. Both exudates and optic disc (OD) are almost identical, so it is necessary to conduct the region segmentation to remove OD while want to detect exudates. To obtain new insight into computer-aided diagnosis (CAD) on DR, this work aims to segment OD and exudates, by implementing the Mask Region-based Convolutional Neural Network (Mask R-CNN) method and the ResNet50 as the backbone network. For the dataset, we used the IDRiD dataset, which contains 54 train images and 27 test images. In this work, we applied Intersection over Union (IoU) as an evaluation parameter. As the result, our proposed scheme gained the IoU value up to 0.8431 and 0.9933 in segmenting OD and exudates on each. It indicates that the applied scheme can support the establishment of CAD for the segmentation of OD and exudates.